In matthiasgomolka/sdcLog: Tools for Statistical Disclosure Control in Research Data Centers

library(xaringanthemer)
library(sdcLog)
library(DiagrammeR)
style_mono_accent(
  base_color = "#1d3557",
  header_font_google = google_font("Fira Sans"),
  text_font_google   = google_font("Fira Sans", "300", "300i"),
  code_font_google   = google_font("Fira Mono")
)

options(htmltools.dir.version = FALSE)
options(datatable.print.keys = FALSE)
options(datatable.print.class = FALSE)

knitr::opts_chunk$set(
  # prompt = TRUE,
  # comment = "#>",
  collapse = TRUE,
  background = "#FAFAFA"
)

Who am I?

And why do I talk about sdcLog?

I work in the Bundesbank's Research Data and Service Centre.

--

What I do:

--

• Production of research data sets on securities transactions
• Data Production Pipelines
• R tools, which make the RDSC life easier

--

Disclaimer:

• No expert in Output Control. I just implemented functionality which we already have for Stata in R.

---

Motivation

Problem

• Researchers need to show that their output complies to our rules.

--

• That get's complicated quickly.

--

• It would be very time-consuming for the RDSC if we would have to check how a researcher proved that her output complies to RDSC rules.

--

Solution

• RDSC provides tools which help researchers to show that their output complies to the rules: sdcLog

---

Theory

Two simple rules:

--

1. Each result must be based on at least 5 distinct entities (distinct ID's).

--

1. The two largest entities must not account for more than 85% of a result (n,k-dominance).

---

Example

A researcher wants to publish the mean of a variable grouped by sector. To do so, she has to use sdc_descriptives() to show that the output complies to RDSC rules.

--

.pull-left[

data("sdc_descriptives_DT")
DT <- sdc_descriptives_DT[, id_na := NULL]
head(DT)

]

--

.pull-right[

# result
DT[, .(mean = mean(val_1, na.rm = TRUE)),
   by = "sector"]

]

--

# Proof, that the result complies to rules
sdc_descriptives(DT, id_var = "id", val_var = "val_1", by = "sector")

---

Another example

This time, researches want to calculate the result grouped by sector and year.

--

sdc_descriptives(DT, id_var = "id", val_var = "val_1", by = c("sector", "year"))

---

Minimum and maximum values

Now, researchers want to publish minimum and maximum values as well.

--

Problem

Minimum and maximum value are confidential micro data.

--

Solution

"Minimum" and "maximum" value as mean of n smallest / largest values using sdc_min_max():

--

sdc_min_max(DT, id_var = "id", val_var = "val_1")

---

Output control for models

Researchers also want to publish results from a linear regression.

--

options(sdc.n_ids = 3)

# Estimate model
mod <- lm(val_1 ~ sector + year + val_2, data = DT)

# Check if model complies to rules
sdc_model(DT, model = mod, id_var = "id")

---

Why is it called sdcLog?

grViz("
digraph boxes_and_circles {

  # a 'graph' statement
  graph [overlap = true, fontsize = 30]

  # several 'node' statements
  node [shape = box, fontname = 'Fira Sans', width = 5.5]
  script [label = 'analysis.R\ncontains sdc_descriptives() / sdc_min_max() / sdc_model()'];

  node [width = 1] 
  log [label = 'Log file']

  node [shape = box, width = 3]
  checks [label = 'Checked by RDSC']

  node [shape = oval, width = 1] 
  sdc_log [label = '@@1']

  # several 'edge' statements
  script -> sdc_log
  sdc_log -> log
  log -> checks
}

[1]: 'sdc_log(analysis.R)'
")

---

Installation und contact information

CRAN

install.packages("sdcLog")

GitHub

https://github.com/matthiasgomolka/sdcLog

E-mail

[e-mail]

Twitter

@matthiasgomolka

matthiasgomolka/sdcLog documentation built on July 17, 2025, 3:21 a.m.